Visa's employer brand team did something most talent teams haven't gotten around to yet. They went into AI search, generated the list of questions they believed candidates were asking about Visa, loaded more than a hundred prompts into their analytics platform, and measured what actually got cited.
The questions about Visa were barely searched.
Almost no one was asking whether Visa has work-life balance. They were asking which financial technology companies have work-life balance — and Visa appeared in that answer only when Visa's content happened to be part of it.
That finding is worth pausing on, because Visa is a globally recognized brand operating in 200 countries. If brand-name search volume is thin for an employer at that scale, it is thin for everyone else too.
This came out of a conversation Dalia hosted with Summer Delaney, founder of CollabWork, and Celinda Appleby, Director of Employer Brand at Visa. The subject was AEO candidate discovery: structuring, hosting, and distributing employer content so that answer engines like ChatGPT, Gemini, and Perplexity can find it, understand it, and cite it when a candidate asks. It is the recruiting application of what marketing teams call AEO (answer engine optimization) or GEO (generative engine optimization).
What follows is what the data shows, why most employer brand content is invisible to these systems, and the specific changes that move the needle fastest.
Key Takeaways
Candidates are searching the category, not the company. Visa's testing found brand-specific prompts were far less common than industry-level questions like "which fintech companies have work-life balance."
Your career site is often not what gets cited. For Visa, the sources that surfaced were Reddit, Glassdoor, Indeed reviews, and employees' own LinkedIn content.
Career site traffic from LLMs is rising sharply, and job seekers — including hourly candidates — are now contacting employer support teams asking how to use ChatGPT in their search.
Where you host content matters as much as writing it. Visa built an eight-page candidate FAQ, hosted it inside their ATS, and it does not surface in AI search at all.
Restructuring job content produces fast results. One CollabWork client saw a 2X increase in career site traffic and a 2.5X increase in job traffic from LLMs within 30 days.
Generic employer brand language gets skipped. Phrases like "come grow with us" are used by so many companies that they cannot differentiate anyone.
The Assumption Behind Most Employer Brand Content
There is a reasonable assumption built into how employer brand content gets planned: that candidates researching your company will ask questions about your company. Teams build FAQ pages about their culture, their benefits, their interview process. The content answers questions phrased around the brand.
This assumption is increasingly wrong, and it explains why a great deal of well-produced employer brand content never appears in AI search results.
Visa's team ran the test properly. They used Perplexity to generate candidate questions, then loaded over a hundred prompts into BrightEdge — the integrated search platform Visa uses — and ran an A/B comparison between Visa-specific questions and category-level questions covering fintech, payments companies, and technology employers in the Bay Area.
The questions the team had originally written, based on years of employer brand experience, were not the questions candidates were asking. Brand-specific prompts were significantly under-searched. The volume sat at the category level.
Candidates do not begin with the employer. They begin with the industry, the function, the location, or the pay band. The employer enters the conversation only as one possible answer.
What This Means for Your Content
If your employer content only answers questions framed around your company, you are absent from the conversation that is actually happening. The candidate asking which healthcare systems offer tuition reimbursement, or which warehouses in Dallas pay above $22 an hour, will receive an answer. That answer will name employers. Whether it names you depends on whether your content addresses the category question, not just the brand question.
This reframing changes what employer brand teams should be producing. Content organized entirely around "what it's like to work here" competes for search volume that is smaller than most teams assume. Content that positions the company within an industry, a role type, or a geography competes for the volume that exists.
The Traffic Shift Behind This
Dalia sits on career sites for high-volume employers and provides direct support to job seekers, which produces two useful signals.
The first is support volume. Job seekers began contacting support with questions about information they had found on ChatGPT, and asking for guidance on how to use it more effectively in their search. A significant portion of these are hourly candidates. When a behavior shift appears in inbound support tickets, it has moved well beyond early adopters.
The second is referral data. The share of career site traffic arriving from ChatGPT is substantially higher than it was a year ago. That traffic clusters around questions about the company and what it is like to work there. Employers with content answering those questions see the traffic. Employers without it do not — the model answers from another source.
Discovery Now Happens Before the Click
There is a measurement problem worth understanding before building a business case around this.
AI search is producing a zero-click dynamic. A candidate asks ChatGPT about your company, receives a useful answer, continues the conversation with the model, and does not click through in that session. Research cited during the conversation found that roughly 60% of users who had a ChatGPT interaction visited the relevant site about a week later.
The practical consequence is that candidates arrive at your career site days after the conversation that actually influenced them, and your analytics record the visit as organic or direct. The discovery happened in the model. The credit goes somewhere else.
Two things follow from this. Your LLM-influenced traffic is almost certainly larger than your attribution reports show. And teams that want to act on this now will need to accept imperfect measurement in the near term.
What Actually Gets Cited
The second finding from Visa's testing was as significant as the first. What surfaced for Visa was largely not the career site. It was reviews on Reddit, Glassdoor, and Indeed, along with LinkedIn content published by individual employees.
Third-party sources and human voices outranked owned brand pages.
This is consistent with what these models appear to reward. Content published by identifiable people, containing specific and unhedged perspectives, is cited more readily than corporate pages. Summer pointed to research showing that LinkedIn posts with sharp, specific opinions were cited more often than safer posts with considerably higher engagement.
For employer brand teams, this means the owned-versus-earned question is no longer a strategic choice. Both require active management, because both are being read.
Six Ways to Improve AEO Candidate Discovery
1. Answer the Questions Every Job Seeker Asks
The most common question Dalia receives from job seekers, consistently and across employers, is some version of what is the status of my application? Almost no employer has published a page answering it.
Dalia began publishing those pages directly — a straightforward article explaining how to check application status at a given company — and they draw meaningful traffic from AI search. Not because the content is sophisticated, but because it fills a void that nearly every employer has left open.
The same applies to "how do I apply." It appears too obvious to document. It is asked constantly, and employers who answer it get surfaced.
This is the fastest available win, and it requires an afternoon.
2. Host Your Content Where AI Can Read It
Visa's talent acquisition team built an eight-page candidate FAQ covering everything from pre-application through offer and onboarding. It is genuinely strong work. It was hosted on SmartRecruiters, then moved to Workday.
It does not surface in AI search.
Content hosted inside an ATS is effectively invisible to answer engines. If your best candidate-facing content lives behind your applicant tracking system, the investment in producing it has not yet reached the audience it was written for. Getting it onto your own domain is the step that makes it count.
3. Lead With the Role, Not the Company
Most job descriptions open with a paragraph about the organization. That order should be reversed.
The first sentence should describe the job: what the person does, in which unit, for what pay. Company context belongs after that, not before it.
Internal shorthand should be removed entirely. A title like "RN2 DOU" communicates nothing to a model or to a candidate — "registered nurse in the direct observation unit" does. The same applies to internal level designations and abbreviated shift codes that were never meant to be public.
Pay belongs near the top rather than at the bottom of the posting. The average prompt entered into an LLM runs roughly 23 words, compared with about three for a traditional Google search. Candidates are specifying role, location, hours, and pay in a single request. Content that does not answer at that level of detail does not enter the response.
This is where results appear fastest. One CollabWork client restructured their job content and XML feed and recorded a 2X increase in career site traffic and a 2.5X increase in job traffic from LLMs over 30 days — from restructuring alone, before any additional site or community work.
4. Replace Generic Language With a Specific Point of View
Employer brand messaging built on phrases like "come grow with us" or "build a future with us" is competing against more than a hundred companies using identical language, including Google and Microsoft. That competition cannot be won on their own words.
The alternative is specificity. Brex ran a campaign built around how many employees had left the company to start their own businesses — not a claim about growing careers, but a countable claim about producing founders. Months later, asking an LLM what it is like to work at Brex surfaced that positioning directly. The message was specific enough to survive summarization.
A useful test: does your messaging indicate who would not thrive at your company? Messaging that applies equally to every employer will not be cited for any of them.
5. Check What Your Site Is Blocking
Many employers have blocked AI crawlers, generally as a defense against mass-apply tools. The unintended consequence is that answer engines cannot read anything the employer publishes.
Review your robots.txt for GPTBot, Google-Extended, ClaudeBot, and PerplexityBot specifically. Blocking discovery in order to prevent application spam addresses the problem in the wrong place. Automated application volume is better handled at the ATS layer, where a text shortcode blocks the large majority of it, while your content remains discoverable.
6. Follow the Citations
The characteristic that makes this manageable is that answer engines disclose their sources.
There is no need to rewrite an entire career site or monitor every platform simultaneously. Ask the models the questions your candidates ask, review which sources are cited, and begin with those specific pages and threads. In many cases a Reddit thread several years old is doing more to shape an employer's reputation than anything the company has published recently.
Data most teams already hold can direct this work: Google Search Console for existing organic performance, interview transcripts for the questions candidates ask live, candidate experience survey responses, and support ticket history. That is a content roadmap available at no cost.
Video is worth adding where budget allows. YouTube transcripts are unusually rich source material for these models, and Gemini draws on YouTube heavily. Visa's integrated search team advised that video would deliver results faster than career site updates. Content already being produced should be uploaded with transcripts attached.
Why This Investment Compounds
This category is genuinely early. Twelve months ago it was not a significant topic at industry conferences. It now represents a majority of what employer brand teams report thinking about. Traffic volumes remain modest relative to established channels, and the full return — particularly whether these candidates produce better hires — is still being established.
The mechanics, however, favor early movers. Once a company becomes an established authoritative source about itself, that position is durable. Paid placement is also coming to these platforms, and building organic presence now costs a fraction of what buying visibility will cost later. Talent acquisition teams have watched this pattern play out with job boards already.
The balance to strike is between short-term hiring reality and long-term positioning. Volumes today are small enough that this cannot replace current sourcing. They are also compounding quickly enough that waiting for the volume to justify the work means paying considerably more to catch up.
Rethink Where Your Employer Brand Lives
Employer brand has historically been treated as something a company publishes. AI search has made it something a company participates in.
The questions candidates ask are broader than any single brand. The sources that answer them are largely not owned. And the answer is being assembled and delivered whether or not the employer contributed to it.
As Summer put it during the conversation: if you are not shaping the answer, someone else is shaping it for you. At present, for most employers, that someone is a several-year-old forum thread.
See Where You Stand
The fastest way to understand your position is to look at it directly. Open ChatGPT, ask the five questions your candidates ask most, and record which sources it cites. That exercise takes twenty minutes and will tell you more about your current employer brand than most quarterly reports.
To understand how much of your career site traffic is already originating in AI search, Dalia can measure it. Dalia sits on career sites for high-volume employers and tracks this shift daily across comparable organizations.
Get your career site AI traffic benchmark →
Thanks to Summer Delaney (CollabWork) and Celinda Appleby (Visa) for the conversation this piece is based on.